I am Zhipei Xu (徐志沛), a second-year Master’s student at the School of Electronic and Computer Engineering, Peking University, advised by Prof. Jian Zhang. Previously, I received my B.Eng degree from the School of Electronic and Information Engineering, South China University of Technology. Please feel free to reach out via email (zhipeixu@stu.pku.edu.cn).

My research focuses on Trustworthy Multi-modal AI, with specific interests in Multi-modal Large Language Models, Image/Video Forgery Detection, and AIGC Security. I have published multiple papers at top-tier venues including CVPR, ICLR, NeurIPS, and ACM MM, with works spanning explainable forgery localization, multi-agent detection frameworks, and robust watermarking/steganography for copyright protection. For more details on my publications, please visit my profiles on Google Scholar ().

🔥 News

• 2025.08.02:  🎉🎉 Gaussianseal has been accepted by MIR!
• 2025.02.27:  🎉🎉 OmniGuard have been accepted by CVPR 2025!
• 2025.01.23:  🎉🎉 FakeShield and SecureGS has been accepted by ICLR 2025!

📝 Selected Publications

(* Equal contribution, † Project Leader, ‡ Corresponding author)

ICLR 2025

FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models
Zhipei Xu, Xuanyu Zhang, Runyi Li, Zecheng Tang, Qing Huang, Jian Zhang
International Conference on Learning Representations (ICLR), 2025

Project Page | Data & Code

• We propose the explainable IFDL task and design FakeShield, a multi-modal framework capable of evaluating image authenticity, generating tampered region masks, and providing a judgment basis based on pixel-level and image-level tampering clues.

Under Review

AvatarShield: Visual Reinforcement Learning for Human-Centric Synthetic Video Detection
Zhipei Xu, Xuanyu Zhang, Qing Huang, Xing Zhou, Jian Zhang
Under Review

• We focus on pose, audio, and text-driven human video forgery and propose the first human-centered video forgery dataset, FakeHumanVid, along with the first reinforcement learning-based human video forgery detection framework, AvatarShield.

Under Review

UniShield: An Adaptive Multi-Agent Framework for Unified Forgery Image Detection and Localization
Qing Huang*, Zhipei Xu*, Xuanyu Zhang, Xiangyu Yu, Jian Zhang
Under Review

• We propose the first multi-agent framework, UniShield, designed to address image forgery detection tasks. It efficiently utilizes a Perception Agent and a Detection Agent to call upon various expert detectors, enabling unified image forgery detection, including DeepFake, AIGC, image forgery, and document forgery.

🎖 Selected Honors and Awards

• 2023.10 National Scholarship.
• 2023.10 National College Student Mathematical Modeling Competition - Provincial First Prize.
• 2022.10 National Scholarship.
• 2021.10 Samsung Scholarship.

📖 Educations

• 2024.09 - 2027.06 (expected), MSc., Computer Science and Technology, Peking University.
• 2020.09 - 2024.06, B.S., Communication Engineering, South China University of Technology.

💬 Services

• Conference Reviewer for ICLR, ICML, NeurIPS, ECCV, ICCV, ACM MM.
• Journal Reviewer for IEEE TIP, IEEE TCSVT, VCIP.